I was trying to build a FAQ bot and since I have many questions to handle, I was thinking of a way to help model, And I am asking if it’s possible to use multiple categories under the main faq intent, and this is just to help the model make predictions on the right category,
for example: I can ask the user to select one of my 10 categories, let’s say I am building a hotel bot, one user may want to ask about reservation timing, so I can show him a list of buttons where he can just select the category “Reservation” then he can ask about the timing since we already know the category it will be easy for the model to get the right answer!
the problem that I have is I don’t know how to add these categories?
is creating multiple faq intents will do the trick for example: faq_reservation, faq_rooms, faq_pricing … and adding multiple ResponseSelector ?
hey @zack, can you help me understand a bit more? Are you looking to add context to the conversation by the bot knowing they are in the “reservation” category from a previous utterance, or are you just looking for ways to improve intent classification?
hi @desmarchris,
Actually I am looking to improve my intent classification on FAQ, and splitting these FAQs to multiple categories is something that I think will to the trick, because my main focus will be on FAQs and I have a domain specific usecase where FAQs intents are so close !
Oh ok. So you’re mainly worried about intent confusion? Can you provide some examples? Usually it’s best to pull this out of the response selector and add entities to help you determine the right answer, but it depends on the use case.
Hi @desmarchris, I’m facing the same issue as zack: I have 120 FAQS (and growing) and some of them are very similar. In that case, what would you do to improve intent classification? By “to pull this out of the response selector” you mean writing the conflictive ones as normal intents with entities instead of retrieval intents?
Oh the other hand, I like the idea of classifying faqs in different categories so the bot could know about which topic is the user talking about. In my case, all FAQS responses will always be the same (no context) but I want to add some type of context to improve the overall experience: suggesting new topics (and keeping them alive if user is interested), provide relevant info based on user preferences etc.
So, what could be the best approach for this situation? Retrieval intents for FAQS + normal intents for contextual topic suggestions?
I think so. It’s a bit abstract talk now without seeing examples but from what I understand, that’s what I would try. Then you can provide the contextual text by branching with slots.